ArchiveBox/archivebox/extractors/__init__.py

189 lines
7.1 KiB
Python

__package__ = 'archivebox.extractors'
import os
from pathlib import Path
from typing import Optional, List, Iterable, Union
from datetime import datetime, timezone
from django.db.models import QuerySet
from ..index.schema import Link
from ..index.sql import write_link_to_sql_index
from ..index import (
load_link_details,
write_link_details,
)
from ..util import enforce_types
from ..logging_util import (
log_archiving_started,
log_archiving_paused,
log_archiving_finished,
log_link_archiving_started,
log_link_archiving_finished,
log_archive_method_started,
log_archive_method_finished,
)
from ..search import write_search_index
from .title import should_save_title, save_title
from .favicon import should_save_favicon, save_favicon
from .wget import should_save_wget, save_wget
from .singlefile import should_save_singlefile, save_singlefile
from .readability import should_save_readability, save_readability
from .mercury import should_save_mercury, save_mercury
from .pdf import should_save_pdf, save_pdf
from .screenshot import should_save_screenshot, save_screenshot
from .dom import should_save_dom, save_dom
from .git import should_save_git, save_git
from .media import should_save_media, save_media
from .archive_org import should_save_archive_dot_org, save_archive_dot_org
from .headers import should_save_headers, save_headers
def get_default_archive_methods():
return [
('title', should_save_title, save_title),
('favicon', should_save_favicon, save_favicon),
('headers', should_save_headers, save_headers),
('singlefile', should_save_singlefile, save_singlefile),
('pdf', should_save_pdf, save_pdf),
('screenshot', should_save_screenshot, save_screenshot),
('dom', should_save_dom, save_dom),
('wget', should_save_wget, save_wget),
('readability', should_save_readability, save_readability), # keep readability below wget and singlefile, as it depends on them
('mercury', should_save_mercury, save_mercury),
('git', should_save_git, save_git),
('media', should_save_media, save_media),
('archive_org', should_save_archive_dot_org, save_archive_dot_org),
]
ARCHIVE_METHODS_INDEXING_PRECEDENCE = [('readability', 1), ('singlefile', 2), ('dom', 3), ('wget', 4)]
@enforce_types
def ignore_methods(to_ignore: List[str]):
ARCHIVE_METHODS = get_default_archive_methods()
methods = filter(lambda x: x[0] not in to_ignore, ARCHIVE_METHODS)
methods = map(lambda x: x[0], methods)
return list(methods)
@enforce_types
def archive_link(link: Link, overwrite: bool=False, methods: Optional[Iterable[str]]=None, out_dir: Optional[Path]=None) -> Link:
"""download the DOM, PDF, and a screenshot into a folder named after the link's timestamp"""
# TODO: Remove when the input is changed to be a snapshot. Suboptimal approach.
from core.models import Snapshot, ArchiveResult
try:
snapshot = Snapshot.objects.get(url=link.url) # TODO: This will be unnecessary once everything is a snapshot
except Snapshot.DoesNotExist:
snapshot = write_link_to_sql_index(link)
ARCHIVE_METHODS = get_default_archive_methods()
if methods:
ARCHIVE_METHODS = [
method for method in ARCHIVE_METHODS
if method[0] in methods
]
out_dir = out_dir or Path(link.link_dir)
try:
is_new = not Path(out_dir).exists()
if is_new:
os.makedirs(out_dir)
link = load_link_details(link, out_dir=out_dir)
write_link_details(link, out_dir=out_dir, skip_sql_index=False)
log_link_archiving_started(link, out_dir, is_new)
link = link.overwrite(updated=datetime.now(timezone.utc))
stats = {'skipped': 0, 'succeeded': 0, 'failed': 0}
for method_name, should_run, method_function in ARCHIVE_METHODS:
try:
if method_name not in link.history:
link.history[method_name] = []
if should_run(link, out_dir, overwrite):
log_archive_method_started(method_name)
result = method_function(link=link, out_dir=out_dir)
link.history[method_name].append(result)
stats[result.status] += 1
log_archive_method_finished(result)
write_search_index(link=link, texts=result.index_texts)
ArchiveResult.objects.create(snapshot=snapshot, extractor=method_name, cmd=result.cmd, cmd_version=result.cmd_version,
output=result.output, pwd=result.pwd, start_ts=result.start_ts, end_ts=result.end_ts, status=result.status)
# bump the updated time on the main Snapshot here, this is critical
# to be able to cache summaries of the ArchiveResults for a given
# snapshot without having to load all the results from the DB each time.
# (we use {Snapshot.id}-{Snapshot.updated} as the cache key and assume
# ArchiveResults are unchanged as long as the updated timestamp is unchanged)
snapshot.save()
else:
# print('{black} X {}{reset}'.format(method_name, **ANSI))
stats['skipped'] += 1
except Exception as e:
raise Exception('Exception in archive_methods.save_{}(Link(url={}))'.format(
method_name,
link.url,
)) from e
# print(' ', stats)
try:
latest_title = link.history['title'][-1].output.strip()
if latest_title and len(latest_title) >= len(link.title or ''):
link = link.overwrite(title=latest_title)
except Exception:
pass
write_link_details(link, out_dir=out_dir, skip_sql_index=False)
log_link_archiving_finished(link, link.link_dir, is_new, stats)
except KeyboardInterrupt:
try:
write_link_details(link, out_dir=link.link_dir)
except:
pass
raise
except Exception as err:
print(' ! Failed to archive link: {}: {}'.format(err.__class__.__name__, err))
raise
return link
@enforce_types
def archive_links(all_links: Union[Iterable[Link], QuerySet], overwrite: bool=False, methods: Optional[Iterable[str]]=None, out_dir: Optional[Path]=None) -> List[Link]:
if type(all_links) is QuerySet:
num_links: int = all_links.count()
get_link = lambda x: x.as_link()
all_links = all_links.iterator()
else:
num_links: int = len(all_links)
get_link = lambda x: x
if num_links == 0:
return []
log_archiving_started(num_links)
idx: int = 0
try:
for link in all_links:
idx += 1
to_archive = get_link(link)
archive_link(to_archive, overwrite=overwrite, methods=methods, out_dir=Path(link.link_dir))
except KeyboardInterrupt:
log_archiving_paused(num_links, idx, link.timestamp)
raise SystemExit(0)
except BaseException:
print()
raise
log_archiving_finished(num_links)
return all_links